Climate change is making these blooms more frequent and unpredictable, putting additional pressure on water management and environmental protection efforts.
Faster detection for safer waters
Researchers at LIST have developed an innovative solution that combines ground-based technology with advanced artificial intelligence to detect harmful cyanobacteria blooms in near real-time. By deploying camera traps around the lake, images of the water surface are captured automatically every hour and sent to LIST researchers for immediate analysis. This allows for a much more detailed and timely understanding of bloom dynamics than traditional monitoring methods, which often rely on occasional water sampling or satellite images limited by cloud cover and revisit schedules.
AI that learns to spot trouble
The system relies on two key components: automated photo traps and machine learning. The cameras, positioned at strategic points, continuously photograph the lake surface. These images are analyzed using YOLO (You Only Look Once), a state-of-the-art object detection algorithm. AI identifies and locates cyanobacteria blooms, providing actionable insights in near real-time.
To train AI, the researchers created a carefully annotated dataset of over a thousand images taken at different times of the year and from multiple locations. Each bloom was precisely marked, and AI learned to distinguish harmful algae from reflections, shadows, and other visual noise. Explainable AI tools like Grad-CAM highlight what the system “sees,” helping researchers refine predictions and reduce errors.
Early warnings for better decisions
This automated monitoring approach enables early warning systems that benefit public authorities, water managers, and lake operators alike. Authorities can act quickly to protect public health, while industrial and recreational stakeholders gain insights for safe and responsible water use.
With additional cameras now deployed and thousands of new images being collected, the AI models are continually improving. These datasets allow researchers to map bloom patterns more precisely, which can then inform predictive models. This forward-looking approach means lake managers can anticipate blooms, rather than reacting after they occur, supporting long-term environmental and operational planning.
This article is based on “Fully Automated Detection of Harmful Cyanobacteria Blooms in Lakes Using Photo Traps and Machine Learning,” by Jean-Baptiste Burnet and Olivier Parisot, published in ERCIM News, special theme article, p. 33. Full article: https://ercim-news.ercim.eu/images/stories/EN143/EN143-web.pdf
